I. INTRODUCTION II. RELATED WORK. International Journal of Computer Science Trends and Technology (IJCST) Volume 5 Issue 5, Sep Oct 2017

Transcription

1 RESEARCH ARTICLE OPEN ACCESS Gradient Histogram Estimation and Preservation for texture Enhanced Image Denoising Guntupalli Sravan Chand [1], Shaik Rahamtula [2] P.G. Student [1], Associate Professor [2] Department of Electronics and Communication Engineering QISIT, Ongole India ABSTRACT Typical snapshot information performs an main position in photograph denoising, and various usual picture priors, including gradient-centered, sparse illustration-situated, and nonlocal self similarity- situated ones, have been extensively studied and exploited for noise removing. In spite of the great success of many denoising algorithms, they have a tendency to soft the high-quality scale image textures when taking away noise, degrading the snapshot visual exceptional. To deal with this crisis, on this paper, we endorse a texture more desirable picture denoising process through implementing the gradient histogram of the denoised image to be just about a reference gradient histogram of the long-established snapshot. Given the reference gradient histogram, a novel gradient histogram renovation (GHP) algorithm is developed to enhance the texture buildings while casting off noise. Two neighborhood-founded editions of GHP are proposed for the denoising of pictures including areas with one-of-a-kind textures. An algorithm is also developed to conveniently estimate the reference gradient histogram from the noisy remark of the unknown snapshot. Our experimental outcome display that the proposed GHP algorithm can good retain the feel looks within the denoised graphics, making them appear more normal. Keywords:- Image denoising, histogram specification, non-local similarity, sparse representation. I. INTRODUCTION Snapshot denoising, which targets to estimate the latent clean photo x from its noisy observation y, is a classical but nonetheless energetic matter in picture processing and low stage vision. One widely used data observation mannequin [4], [7], [9] [11] is y = x+v, the place v is additive white Gaussian noise (AWGN). One general procedure to photograph denoising is the variational approach, where an vigor useful is minimized to go looking the favored estimation of x from its noisy statement y. The vigour practical traditionally involves two phrases: a data constancy term which is dependent upon the image degeneration process and a regularization time period which items the prior of unpolluted ordinary pix [4], [7], [8], [12]. The statistical modeling of natural Photo priors are principal to the success of picture denoising. Prompted with the aid of the truth that normal image gradients and wavelet turn out to be coefficients have a heavy-tailed distribution, sparsity priors are commonly used in photo denoising [1] [3]. The well-known total variant minimization approaches clearly assume Laplacian distribution of photo gradients [4]. The sparse Laplacian distribution can also be used to mannequin the excessive-go filter responses and wavelet/curvelet turn out to be Coefficients [5], [6]. Via representing picture patches as a sparse linear mixture of the atoms in an overwhole redundant dictionary, which will also be analytically designed or realized from ordinary portraits, sparse coding has proved to be very mighty in image denoising via l0-norm or l1-norm minimization [7], [8]. Yet another general prior is the nonlocal self-similarity (NSS) prior [9] [11], [50]; that's, in usual graphics there are usually many similar patches (i.e., nonlocal neighbors) to a given patch, which may be spatially far from it. The connection between NSS and the sparsity prior is mentioned in [11] and [12]. The joint use of sparsity prior and NSS prior has resulted in ultra-modern photo denoising results [12] [14]. In spite of the quality success of many denoising Algorithms, nevertheless, they traditionally fail to preserve the image satisfactory scale texture constructions [23], degrading a lot the photo visible exceptional. II. RELATED WORK Photograph denoising ways can also be grouped into two classes: Mannequin-founded methods and studying-founded approaches. M Most denoising methods reconstruct the clean image bymexploiting some image and noise prior items, and belong to the first class. Learning-founded methods attempt tostudy a mapping operate from the noisy snapshot to the clean image [19], and have been receiving tremendous research interests [20], [21]. Right here we in short evaluate those model-situated denoising approaches concerning our work from a perspective of natural photo priors experiences on common image priors ISSN: Page 81

2 aim to search out suitable items to describe the characteristics or information (e.g., distribution) of photographs in some domain. One representative classification of photograph priors is the gradient prior situated on the remark that natural graphics have a heavy-tailed distribution of gradients. Using gradient prior may also be traced again to Nineties when Rudin et al. [4] proposed a complete variant (television) model for image denoising, the place the gradients are really modeled as Laplacian distribution. One other famous prior mannequin, themixture of Gaussians, will also be used to approximate the distribution of photo gradient [1], [22]. In addition, hyper- Laplacian mannequin can extra effectively symbolize the heavy tailed distribution of gradients, and has been commonly applied to quite a lot of snapshot restoration tasks [2], [3], [23] [25]. The photo gradient prior is a sort of local sparsity prior, i.e., the gradient distribution is sparse. More most commonly, the neighborhood sparsity prior can also be good applied to excessive-pass filter responses, wavelet/curvelet turn into coefficients, or the coding coefficients over a redundant dictionary. In [5] and [6], Gaussian scale combinations are used to symbolize the marginal and joint distributions of wavelet grow to be coefficients. Via assuming that an image patch can be represented as a sparse linear combination of the atoms in an over-entire dictionary, a quantity of dictionary finding out (DL) approaches (e.g., evaluation and synthesis k-svd [7], [28], task pushed DL [29], and adaptive sparse domain resolution [8] were proposed and applied to picture denoising and different image restoration tasks. Established on the truth that a identical patch to the given patch may not be spatially virtually it, a different line of study is to model the similarity between image patches, i.e., the picture nonlocal self-similarity (NSS) priors. The seminal work of nonlocal way denoising [9] has encouraged a extensive variety of studies on NSS, and has led to a flurry of NSS founded state Of-the-artwork denoising ways, e.g., BM3D [11], LSSC [12], and EPLL [30], and many others. Noised Image Estimation Gradient Histogram Preserving Iterative Histogram Specification Gradient Histogram Denoised Image Advanced Learned Simultaneously Sparse Coding Enhanced Output Image Fig:-Flowchart of the proposed texture enhanced image denoising framework.. III. THE TEXTURE ENHANCED IMAGE DENOISING FAME WORK The noisy commentary y of an unknown smooth picture x is most likely modeled as y = x + v,----(1) where v is the additive white Gaussian noise (AWGN) with zero imply and usual deviation σ. The intention of photograph denoising is to estimate the favored photograph x from y. One preferred strategy to photo denoising is the variational process, wherein the denoised snapshot is received through ˆx = argmin x { 1/2 σ2 (y x)2 + λ R(x)} Where R(x) denotes some regularization time period and λ is a confident regular. The targeted form of R(x) depends upon the employed photo priors. IV. DENOISING WITH GRADIENT HISTOGRAM PRESERVATION A. The Denoising Model The proposed denoising approach is a patch situated method. Let xi = Ri x be a patch extracted at position i, i = 1, 2,..., N, where Ri is the patch extraction operator and N is the number of pixels in the photograph. Given a dictionary D, we sparsely encode the patch xi over D, leading to a sparse coding vector αi. As soon as the coding vectors of all photograph patches are received, the entire picture x can be reconstructed with the aid of X=D o α The place α is the concatenation of all αi. 1 Input :- Binary image Output :- Denoised image Algorithm 1 : Step 1 :- Take input image binary image as A Step 2 :-Apply some noise to image as B Step 3:-Applying Gradient histogram method f(x)=aub. Step 4 :- Applying Advanced proposed method advanced_learned_simultaneously_sparse_coding (EXTENSION) ISSN: Page 82

3 Step5:- Result image with enhanced image.. V. REFERENCES GRADIENT HISTOGRAM ESTIMATION To use the model in Eq. (7), we have got to comprehend the reference gradient histogram hr of customary snapshot x. In this section, we propose a regularized deconvolution model to estimate the histogram hr. Assuming that the pixels in gradient photo x are unbiased and identically allotted (i.i.d.), we can view them because the samples of a scalar variable, denoted via x. Then the normalized histogram of x will also be viewed as a discrete approximation of the chance density operate (PDF) of x. For the AWGN v, we will effortlessly model its elements because the samples of an i.i.d. Variable, denoted via v. Since v N 0, σ2 and let ε = v, ε can then be good approximated by means of the i.i.d. Gaussian with PDF pε = 1/2 πσ exp(-ε2/ 4σ2) Since y = x + v, we have y = x + v. It is ready to model y as an i.i.d. variable, denoted by y, and we have y = x +ε. Let px be the PDF of x, and py be the PDF of y. Since x and ε are independent, the joint PDF p (x, ε) is p(x, ε) = px pε.. VI. EXPERIMENTAL RESULTS To verify the efficiency of the proposed GHP established photograph denoising approach, we apply it to ten normal portraits with various texture structures. All the scan photographs are gray-scale portraits with gray level ranging from zero to 255. We first talk about the parameter setting in our GHP algorithm, and then compare the efficiency of international founded GHP and its area headquartered editions, i.e., B-GHP and S-GHP. In the end, experiments are performed to validate its performance in assessment with the state-of-the-art denoising algorithms. Within the following experiments we set the AWGN commonplace deviation from 20 to forty with step length (a) Input image (b) Process Image (c) Result image TABLE Existing method Proposed method Images PSNR SSIM PSNR SSIM vehicle building Tree Animal Person The PSNR (Db) And SSIM Results VII. CONCLUSION Type On this paper, we presented a novel gradient histogram renovation (GHP) model for texture more ISSN: Page 83

4 suitable image denoising, and additional introduce two vicinity-founded GHP variations, i.e., B-GHP and S-GHP. A easy but theoretically solid model and the associated algorithm have been presented to estimate the reference gradient histogram from the noisy image, and an efficient iterative histogram specification algorithm was developed to put in force the GHP mannequin. By using pushing the gradient histogram of the denoised picture towards the reference histogram, GHP achieves promising outcome in bettering the texture structure while removing random noise. The experimental outcome validated the effectiveness of GHP in texture better picture denoising. GHP results in similar PSNR/SSIM measures to the modern day denoising ways equivalent to SAPCABM3D, LSSC and NCSR; however, it results in more ordinary and visually great denoising results by way of higher preserving the snapshot texture areas. Most of the latest denoising algorithms are founded on the nearby sparsity and nonlocal selfsimilarity priors of ordinary photos. Not like them, the gradient histogram utilized in our GHP process is a type of world prior, which is adaptively estimated from the given noisy snapshot. One challenge of GHP is that it can't be directly applied to non-additive noise removing, such as multiplicative Poisson noise and sign-stylish noise [47]. Hence, it might be exciting and priceless to study extra basic units and algorithms for non-additive noise elimination with texture enhancement. One strategy is to transform the noisy photo into an picture with additive white Gaussian noise (AWGN) and then practice GHP. For example, for picture with Poisson noise, Anscombe root transformation [48], [49] can be used to convert it into an photograph with AWGN.. REFERENCES [1] R. Fergus, B. Singh, A. Hertzmann, S. Roweis, and W. T. Freeman, Removing camera shake from a single photograph, in Proc. ACM SIGGRAPH, 2006, pp [2] A. Levin, R. Fergus, F. Durand, and W. T. Freeman, Image and depth from a conventional camera with a coded aperture, in Proc. ACM SIGGRAPH, [3] D. Krishnan and R. Fergus, Fast image deconvolution using hyper- Laplacian priors, in Proc. NIPS, 2009, pp [4] L. Rudin, S. Osher, and E. Fatemi, Non linear total variation based noise removal algorithms, Phys. D, vol. 60, nos. 1 4, pp , Nov [5] M. Wainwright and S. Simoncelli, Scale mixtures of Gaussians and the statistics of natural images, in Proc. NIPS, vol , pp [6] J. Portilla, V. Strela, M. J. Wainwright, and E. P. Simoncelli, Image denoising using a scale mixture of Gaussians in the wavelet domain, IEEE Trans. Image Process., vol. 12, no. 11, pp , Nov [7] M. Elad and M. Aharon, Image denoising via sparse and redundant representations over learned dictionaries, IEEE Trans. Image Process., vol. 15, no. 12, pp , Dec [8] W. Dong, L. Zhang, G. Shi, and X. Wu, Image deblurring and superresolution by adaptive sparse domain selection and adaptive regularization, IEEE Trans. Image Process., vol. 20, no. 7, pp , Jul [9] A. Buades, B. Coll, and J. Morel, A review of image denoising methods, with a new one, Multiscale Model. Simul., vol. 4, no. 2, pp , [10] K. Dabov, A. Foi, V. Katkovnik, and K. Egiazarian, Image denoising by sparse 3D transform-domain collaborative filtering, IEEE Trans. Image Process., vol. 16, no. 8, pp , Aug [11] V. Katkovnik, A. Foi, K. Egiazarian, and J. Astola, From local kernel to nonlocal multiplemodel image denoising, Int. J. Comput. Vis., vol. 86, no. 1, pp. 1 32, Jan [12] J. Mairal, F. Bach, J. Ponce, G. Sapiro, and A. Zisserman, Non-local sparse models for image restoration, in Proc. Int. Conf. Comput. Vis., Sep./Oct. 2009, pp [13] W. Dong, L. Zhang, and G. Shi, Centralized sparse representation for image restoration, in Proc. Int. Conf. Comput. Vis., Nov. 2011, pp [14] W. Dong, L. Zhang, G. Shi, and X. Li, Nonlocally centralized sparse representation for image restoration, IEEE Trans. Image Process., vol. 22, no. 4, pp , Apr [15] E. Hadjidemetriou, M. D. Grossberg, and S. K. Nayar, Multiresolution histograms and their use for recognition, IEEE. Trans. Pattern Anal. Mach. Intell., vol. 26, no. 7, pp , Jul [16] M. Varma and A. Zisserman, A statistical approach to texture classification from single images, Int. J. Comput. Vis., vol. 62, nos. 1 2, pp , Apr [17] M. Varma and A. Zisserman, Classifying images of materials: Achieving viewpoint and illumination independence, in Proc. ECCV, vol , pp [18] W. Zuo, L. Zhang, C. Song, and D. Zhang, Texture enhanced image denoising via gradient histogram preservation, in Proc. Int. Conf. CVPR, 2013, pp [19] K. Suzuki, I. Horiba, and N. Sugie, Efficient approximation of neural filters for removing ISSN: Page 84

5 quantum noise from images, IEEE Trans. Signal Process., vol. 50, no. 7, pp , Jul [20] V. Jain and H. Seung, Natural image denoising with convolutional networks, in Proc. NIPS, 2008, pp [21] H. C. Burger, C. J. Schuler, and S. Harmeling, Image denoising: Can plain neural networks compete with BM3D? in Proc. Int. Conf. CVPR, Jun. 2012, pp [22] A. Levin, Y. Weiss, F. Durand, and W. T. Freeman, Efficient marginal likelihood optimization in blind deconvolution, in Proc. Int. Conf. CVPR, Jun. 2011, pp [23] T. S. Cho, C. L. Zitnick, N. Joshi, S. B. Kang, R. Szeliski, and W. T. Freeman, Image restoration by matching gradient distributions, IEEE. Trans. Pattern Anal. Mach. Intell., vol. 34, no. 4, pp , Apr [24] T. S. Cho, N. Joshi, C. L. Zitnick, S. B. Kang, R. Szeliski, and W. T. Freeman, A contentaware image prior, in Proc. Int. Conf. CVPR, Jun. 2010, pp [25] N. Joshi, C. L. Zitnick, R. Szeliski, and D. Kriegman, Image deblurring and denoising using color priors, in Proc. Int. Conf. CVPR, Jun. 2009, pp [26] M. Welling, G. Hinton, and S. Osindero, Learning sparse topographic representations with products of student-t distributions, in Proc. NIPS, Dec [27] S. Roth and M. J. Black, Fields of experts: A framework for learning image priors, in Proc. Int. Conf. CVPR, vol. 2. Jun. 2005, pp [28] R. Rubinstein, T. Peleg, and M. Elad, Analysis K-SVD: A dictionarylearning algorithm for the analysis sparse model, IEEE Trans. Signal Process., vol. 62, no. 3, pp , Feb [29] J. Mairal, F. Bach, and J. Ponce, Task driven dictionary learning, IEEE. Trans. Pattern Anal. Mach. Intell., vol. 32, no. 4, pp , Apr [30] D. Zoran and Y. Weiss, From learning models of natural image patches to whole image restoration, in Proc. Int. Conf. Comput. Vis., Nov. 2011, pp [31] A. Levin, B. Nadler, F. Durand, and W. T. Freeman, Patch complexity, finite pixel correlations and optimal denoising, in Proc. 12th ECCV, 2012, pp [32] J. Jancsary, S. Nowozin, and C. Rother, Lossspecific training of nonparametric image restoration models: A new state of the art, in Proc.12th ECCV, 2012, pp [33] I. Daubechies, M. Defriese, and C. De Mol, An iterative thresholding algorithm for linear inverse problems with a sparsity constraint, Commun. Pure Appl. Math., vol. 57, no. 11, pp , Nov [34] R. C. Gonzalez and R. E. Woods, Digital Image Processing. Beijing, China: Publishing House Electron. Ind., [35] L. Lin, X. Liu, and S.-C. Zhu, Layered graph matching with composite cluster sampling, IEEE. Trans. Pattern Anal. Mach. Intell., vol. 32, no. 8, pp , Aug [36] L. Lin, X. Liu, S. Peng, H. Chao, Y. Wang, and B. Jiang, Object categorization with sketch representation and generalized samples, Pattern Recognit., vol. 45, no. 10, pp , [37] S. M. Pizer et al., Adaptive histogram equalization and its variations, Comput. Vis. Graph. Image Process., vol. 39, no. 3, pp , Sep [38] J. K. Patel and C. B. Read, Handbook of the Normal Distribution. New York, NY, USA: Marcel Dekker, [39] Z. Wang, A. C. Bovik, H. R. Sheikh, and E. P. Simoncelli, Image quality assessment: From error visibility to structural similarity, IEEE Trans. Image Process., vol. 13, no. 4, pp , Apr [40] J. Bioucas-Dias and M. Figueiredo, Multiplicative noise removal using variable splitting and constrained optimization, IEEE Trans. Image Process., vol. 19, no. 7, pp , Jul [41] M. Afonso, J. Bioucas-Dias, and M. Figueiredo, Fast image recovery using variable splitting and constrained optimization, IEEE Trans. Image Process., vol. 19, no. 9, pp , Sep [42] M. Afonso, J. Bioucas-Dias, and M. Figueiredo, An augmented Lagrangian approach to the constrained optimization formulation of imaging inverse problems, IEEE Trans. Image Process., vol. 20, no. 3, pp , Mar [43] B. Julesz, Textons, the elements of texture perception, and their interactions, Nature, vol. 290, pp , Mar [44] X. Liu and D. Wang, A spectral histogram model for texton modeling and texture discrimination, Vis. Res., vol. 42, no. 23, pp , Oct [45] J. J. Peissig, M. E. Young, E. A. Wasserman, and I. Biederman, The role of edges in object recognition by pigeons, Perception, vol. 34, no. 11, pp , [46] C. L. Zitnick and D. Parikh, The role of image understanding in contour detection, in Proc. IEEE Conf. CVPR, Jun. 2012, pp [47] B. Zhang, J. M. Fadili, and J. L. Starck, Wavelets, ridgelets, and curvelets for Poisson noise removal, IEEE Trans. Image Process., vol. 17, no. 7, pp , Jul [48] F. J. Anscombe, The transformation of Poisson, binomial and negativebinomial data, ISSN: Page 85

6 Biometrika, vol. 35, nos. 3 4, pp , Dec [49] M. Mäkitalo and A. Foi, Optimal inversion of the Anscombe transformation in lowcount Poisson image denoising, IEEE Trans. Image Process., vol. 20, no. 1, pp , Jan [50] L. Zhang, W. Dong, D. Zhang, and G. Shi, Two-stage image denoising by principal component analysis with local pixel grouping, Pattern Recognit., vol. 43, no. 4, pp , Apr Mr.Guntupalli Sravan Chand received his B.Tech degree from the department of Electronics and Communication Engineering, VELAGA NAGESWARARAO COLLEGE OF ENGINEERING (Affliated to JNTU KAKINADA). He is pursuing M.Tech in QIS Institute of Technology, Ongole, AP, India. His current research interests are Digital Image Processing, Digital Signal Processing. Mr. Shaik. Rahamtula is currently working as assistant professor in ECE Department, QIS Institute of technology, Ongole, A.P, India. He received his B. Tech degree in the department of Electronics and Communication Engineering, from KMCET (Affiliated to JNTU Hyderabad). He received his M. Tech from QIS college of Engineering and Technology (Affiliated to JNTU Kakinada). His research interests in the area of Face recognition, Content based image retrieval, Image compression. ISSN: Page 86